elif arguments['activation'].startswith('pnorm'):
do_pnorm = True
pool_size, pnorm_order = parse_two_integers(arguments['activation'])
train_dataset, train_dataset_args = read_data_args(train_data_spec)
valid_dataset, valid_dataset_args = read_data_args(valid_data_spec)
train_sets, train_xy, train_x, train_y = read_dataset(train_dataset, train_dataset_args)
valid_sets, valid_xy, valid_x, valid_y = read_dataset(valid_dataset, valid_dataset_args)
numpy_rng = numpy.random.RandomState(89677)
theano_rng = RandomStreams(numpy_rng.randint(2 ** 30))
log('> ... building the model')
# doesn't deal with dropout
dnn = DNN_SAT(numpy_rng=numpy_rng, theano_rng = theano_rng, n_ins=n_ins,
hidden_layers_sizes=hidden_layers_sizes, n_outs=n_outs,
activation = activation, do_maxout = do_maxout, pool_size = pool_size,
do_pnorm = do_pnorm, pnorm_order = pnorm_order,
max_col_norm = max_col_norm, l1_reg = l1_reg, l2_reg = l2_reg,
ivec_dim = ivec_dim, ivec_layers_sizes = ivec_layers_sizes)
# read the initial DNN
_file2nnet(dnn.sigmoid_layers, filename = si_model_file)
# get the training, validation and testing function for iVecNN
dnn.params = dnn.ivec_params
dnn.delta_params = dnn.ivec_delta_params
log('> ... getting the finetuning functions for iVecNN')
train_fn, valid_fn = dnn.build_finetune_functions(
(train_x, train_y), (valid_x, valid_y),
batch_size=batch_size)
log('> ... learning the iVecNN network')
while (lrate.get_rate() != 0):
cfg_si.init_data_reading(train_data_spec, valid_data_spec)
# parse the structure of the i-vector network
cfg_adapt = NetworkConfig()
# net_split = adapt_nnet_spec.split(':')
# adapt_nnet_spec = ''
# for n in xrange(len(net_split) - 1):
# adapt_nnet_spec += net_split[n] + ':'
# cfg_adapt.parse_config_dnn(arguments, adapt_nnet_spec + '0')
cfg_adapt.parse_config_dnn(arguments, adapt_nnet_spec + ':0')
numpy_rng = numpy.random.RandomState(89677)
theano_rng = RandomStreams(numpy_rng.randint(2**30))
log('> ... initializing the model')
# setup up the model
dnn = DNN_SAT(numpy_rng=numpy_rng,
theano_rng=theano_rng,
cfg_si=cfg_si,
cfg_adapt=cfg_adapt)
# read the initial DNN (the SI DNN which has been well trained)
_file2nnet(dnn.dnn_si.layers, filename=init_model_file)
# get the training and validation functions for adaptation network training
dnn.params = dnn.dnn_adapt.params # only update the parameters of the i-vector nnet
dnn.delta_params = dnn.dnn_adapt.delta_params
log('> ... getting the finetuning functions for iVecNN')
train_fn, valid_fn = dnn.build_finetune_functions(
(cfg_si.train_x, cfg_si.train_y), (cfg_si.valid_x, cfg_si.valid_y),
batch_size=cfg_adapt.batch_size)
log('> ... learning the adaptation network')
cfg = cfg_adapt
while (cfg.lrate.get_rate() != 0):
cfg_si.parse_config_dnn(arguments, si_nnet_spec)
cfg_si.init_data_reading(train_data_spec, valid_data_spec)
# parse the structure of the i-vector network
cfg_adapt = NetworkConfig()
# net_split = adapt_nnet_spec.split(':')
# adapt_nnet_spec = ''
# for n in xrange(len(net_split) - 1):
# adapt_nnet_spec += net_split[n] + ':'
# cfg_adapt.parse_config_dnn(arguments, adapt_nnet_spec + '0')
cfg_adapt.parse_config_dnn(arguments, adapt_nnet_spec + ':0')
numpy_rng = numpy.random.RandomState(89677)
theano_rng = RandomStreams(numpy_rng.randint(2 ** 30))
log('> ... initializing the model')
# setup up the model
dnn = DNN_SAT(numpy_rng=numpy_rng, theano_rng = theano_rng, cfg_si = cfg_si, cfg_adapt = cfg_adapt)
# read the initial DNN (the SI DNN which has been well trained)
_file2nnet(dnn.dnn_si.layers, filename = init_model_file)
# get the training and validation functions for adaptation network training
dnn.params = dnn.dnn_adapt.params # only update the parameters of the i-vector nnet
dnn.delta_params = dnn.dnn_adapt.delta_params
log('> ... getting the finetuning functions for iVecNN')
train_fn, valid_fn = dnn.build_finetune_functions(
(cfg_si.train_x, cfg_si.train_y), (cfg_si.valid_x, cfg_si.valid_y),
batch_size = cfg_adapt.batch_size)
log('> ... learning the adaptation network')
cfg = cfg_adapt
while (cfg.lrate.get_rate() != 0):
# one epoch of sgd training